A Cubic Convolution Interpolation-Based Chroma Subsampling Method for Bayer and RGBW CFA Raw Images

Abstract

The Bayer and RGBW color filter array (CFA) raw images, denoted by I Bayer and I RGB , respectively, have been widely used in consumer markets. In the demosaicking-first compression scheme, chroma subsampling is necessary prior to compressing I RGB and I RGBW . Several linear interpolationbased chroma subsampling methods have been developed for I Bayer , but no nonlinear interpolation-based chroma subsampling methods have targeted the above two CFA image types simultaneously. In this paper, we first propose a nonlinear interpolation-based, namely the cubic convolution interpolation-based (CCIbased), 2×2 block-distortion function for each 2×2 CFA block B CbCr . Next, using the Cauchy-Schwarz inequality, we prove that the proposed block-distortion function is a convex function in the real domain, which further serves as the base of the initial subsampled chroma solution. Then, a CCI-based iterative method is proposed to improve the initial subsampled chroma solution. The results of comprehensive experimental tests using Bayer and RGBW CFA images created from the three RGB full-color datasets, namely IMAX, SCI (screen content images), and CI (classical images), demonstrate that on the Versatile Video Coding (VVC) platform VTM-11.0, the proposed method achieves substantial quality enhancement and quality-bitrate tradeoff merits of the reconstructed Bayer and RGBW CFA images compared with existing chroma subsampling methods. INDEX TERMSBayer color filter array (CFA) image, Chroma subsampling, Convex block-distortion function, Demosaicking, Quality-bitrate tradeoff, Quality enhancement, RGBW CFA image.

Compression for Bayer CFA Images: Review and Performance Comparison

“…Then, a chroma 4:2:0 subsampling method is performed on the chroma image to obtain a subsampled CbCr image whose size is a quarter of the original chroma image . In Section 3.2 , two kinds of chroma 4:2:0 subsampling approaches [ 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 ], namely the Bayer CFA pattern-independent approach and the Bayer CFA pattern-dependent approach in Figure 3 b, will be introduced. Furthermore, based on the subsampled CbCr image and the luma image , a luma modification method is performed on to obtain a modified luma image .…”